Analysis of occupation and control on memory in multilevel fast multipole algorithm

The relation between memory and variable numbers is found, and the effects of the ratio of grid partition dimensions to incident wavelength and the truncation number to memory are presented based on multilevel fast multipole algorithm (MLFMA) and memory which is the primary part of the storage matrix. When incident frequency is invariable and grid partition dimensions decrease, the memory of radiating pattern and receiving pattern is directly proportional to total unknown numbers, and the sparse matrix memory increases by a square ratio to unknown numbers. The whole memory mushrooms under the influence of them as an affirmative result. When grid data are invariable and incident frequency reduces, the sparse matrix memory does not change and the memory of radiating pattern and receiving pattern would decrease with a reduction in frequency, at the same delamination number, while the sparse matrix memory increases by a square ratio with a reduction in delamination numbers. The memory of radiating pattern and receiving pattern does not change and the oscillation of the whole memory is enhanced when the frequency changes by even number times. After improving the precision control of truncation numbers, the calculating precision is enhanced notably, but the memory of radiating pattern and receiving pattern and the whole memory increase correspondingly.